Adaptive optimal stimulus selection in cognitive models using a model averaging approach

Abstract

Stimulus selection based on the maximum Fisher information (MFI) principle enables the efficient estimation of participant parameters of cognitive models with elaborated experimental tasks. However, in typical applications, a single model is assumed to be true, whereby model uncertainty is ignored, causing poor performance of MFI-based stimulus selection. To address this problem, this study proposes the model-averaged MFI stimulus-selection method that simultaneously considers multiple models based on the model averaging framework. Three simulation studies were conducted to investigate the performance of the proposed method. The results indicate that the proposed method performed as well as in the ideal case in which the true model is known, with conventional MFI under a misspecified model performing worse. Thus, they demonstrate the superiority of the proposed approach, because the true model is unknown in reality. The proposed method enables robust stimulus selection, leading to high efficiency in parameter estimation.